ResNet-Driven Automated Identification of Custard Apple Diseases for Sustainable Smart Agriculture

The development of an automated fruit disease detection system is critical to improve agricultural productivity. This study specifically addressed disease detection in custard apple plants by using a deep learning-based classification approach. Using an extensive image database from different regions, including India, Portugal, Thailand, Cuba and the West Indies, the system successfully classified 8226 images of fruit and leaf diseases of custard apple into six different categories: anthracnose, black canker, diplodia rot, leaf spot on fruit, leaf spot on leaf, and mealy bug. Using transfer learning, the system demonstrated strong classification performance, even with images taken in natural environments with complex backgrounds. By analyzing the unique features of the images, the proposed model accurately identified disease symptoms. In addition, evaluation metrics such as classification accuracy (CA), recall, precision, F1 score and confusion matrix underscored the model’s effectiveness, with ResNet standing out as the most efficient architecture, achieving an impressive 99.77% CA. This study demonstrated the potential of the system to significantly improve disease detection in custard apple crops, and offers a promising tool for improving agricultural management.